Quantitative estimates of variance components in “normal value”

Abstract

In measurement of “normal” persons, it is common to discuss the concept of normal value which is expressed in terms of intervals. Statistical properties of such values have, however, been seldom investigated. The purpose of the present paper is in discussing such properties employing the techniques of random effects model of analysis of variance. As an actual example, the urine concentration test was used. The variation of measurement generally consists of measurement error, variation within persons and between persons. The test for the variance components of the urine concentration test indicated that variations within and between persons were significant (p<1%). The point and interval estimation for them revealed that the measurement error was negligibly small, while variations within and between persons were fairly large and approximately of the same order. From these figures it was concluded that the pycnometer method of specific gravity determination was completely satisfactory. Since variation within persons occupied considerable portion of the total variation, it was concluded that the introduction of the concept of individual normal value might not be so effective as to warrant the effort in the urine concentration test. The effects of the several kinds of replication were also discussed. Administration of anti-diuretic hormone was found ineffective in the modified Fishberg method of urine concentration test adopted for this experiment.